sup-tues

8/8/2019 sup-tues

http://slidepdf.com/reader/full/sup-tues 1/5

3A.2 FORECASTING THE SUPER TUESDAY TORNADO OUTBREAK AT THE SPC:

WHY FORECAST UNCERTAINTY DOES NOT NECESSARILY DECREASE AS

YOU GET CLOSER TO A HIGH IMPACT WEATHER EVENT

Jeffry S. Evans*,Corey M. Mead and Steven J. Weiss

Storm Prediction Center, Norman, OK

1. INTRODUCTION

The 5 February 2008 (Super Tuesday) tornadooutbreak ranks as the deadliest tornado outbreakin 23 years for the United States, and the deadliestsince 3-4 April 1974 in the Ohio and TennesseeValleys (Fig. 1). Eighty-four tornadoes occurredduring the course of the outbreak, killing 57 peopledirectly in Arkansas, Kentucky, Tennessee andAlabama (Fig. 2) and injuring hundreds of otherswithin these and surrounding states. The outbreak

also produced widespread wind damage alongwith large hail up to the size of softballs (4.5inches in diameter).

Synoptically, this event was evident up to aweek in advance as large scale features favored apotential outbreak of severe thunderstorms andtornadoes. A deepening middle and upper leveltrough of low pressure was forecast consistentlyby medium-range numerical models (e.g. GFS,GFS Ensemble, ECMWF) over the central UnitedStates around 5 February. Ahead of this system,several days of southerly low level flow out of theGulf of Mexico was expected to support

unseasonable moisture (i.e. surface dew points inexcess of 60 o F) over the lower Mississippi river valley, the Mid South, and lower Ohio andTennessee river valleys. Low level moisture of this quality is a common feature for cool seasontornado outbreaks across the Southern andSoutheastern U.S. (Guyer et al. 2006). Windsaloft and associated shear profiles were forecastto be anomalously strong at all levels with thissystem.

These “synoptically evident” (Doswell et al.1993) events are quite rare with only a few suchoccurrences as clearly apparent in medium range

model guidance each year. More typically,differing model solutions lead to inherently higher forecast uncertainty several days in advance of asevere weather event. Even if model guidance isin general agreement, the coarser resolution of these data sets limit the assessment of any sub-

*Corresponding Author Address: Jeffry S. Evans,Storm Prediction Center, Norman, OK, 73072email: [email protected]

synoptic scale processes that could significantlyimpact event evolution. Consequently, tools suchas climatology and large scale pattern recognitionplay an important role at this point in the forecastprocess.

Fig 1. Ranking of deadliest tornado outbreaks

since 1950 based on direct fatalities. Two time-frame criteria are used; one based on UTC (12z-12z), and another on midnight CST. (Fromwww.spc.noaa.gov)

Fig 2. Tornado tracks and fatality locations for theFebruary 5-6 tornado outbreak. (Fromwww.spc.noaa.gov/wcm)

mailto:[email protected]



8/8/2019 sup-tues


As the forecast valid time nears 12-24 hours inadvance, forecasters are able to examine finer scale details of the atmosphere through the use of mesoscale models and observational data.Typically, this additional information providesinsight into meso and perhaps even storm scaleprocesses that may have a profound impact onhow a severe weather event evolves. When thesedata are consistent with previous forecastexpectations, then the confidence in the forecastincreases with the approach of the event.However, when the short term data either conflictwith the earlier forecasts or support differingsolutions, then forecast confidence can actuallydecrease as one nears the valid time.

2. MEDIUM-RANGE FORECAST

By as early as 31 January 2008, theconsistency between, and run-to-run continuity of,

the medium-range models provided increasedconfidence of an impending significant, severeweather event. As such, the Storm PredictionCenter (SPC) highlighted the potential threat intheir Day 4-8 convective outlook, 6 days inadvance (Fig. 3). Subsequent SPC outlooksincreased both the potential area of severethunderstorms and the significance of theexpected event. Forecaster confidence supporteda “Moderate Risk” in the initial Day 2 outlookissued at 1 am CST 4 February, which included a≥10% probability forecast of significant severethunderstorms (defined as those producing 2" or

larger hail, 65 knot or stronger winds, and EF2+tornadoes). The Moderate Risk was expandedwestward into more of Arkansas in the subsequentDay 2 outlook (Fig 4) issued at 1230 pm CST onthe 4th, which included mention of “long-livedsupercells and possible strong tornadoes.”

3. SHORT-RANGE FORECAST

Late on 4 February, however, an assessmentof observational data and higher resolutionnumerical model output resulted in SPCforecasters becoming more uncertain in how the

event would unfold. The presence of a lead, lowlatitude short wave trough over the lower RioGrande valley in advance of the primary upper-level trough presented added complexity to theforecast. One plausible scenario was that thislead shortwave trough would initiate storms earlyin the day over eastern Texas into westernLouisiana, eventually overturning much of the air mass prior to the peak of the diurnal heating cycle.Another possible scenario was that the presence

Fig 3. Day 6 outlook issued by the SPC on Jan31st valid from 12z Feb 5 - 12z Feb 6. overlaid onsevere reports valid during that time. Tornadoesare red dots, wind damage blue and hail green.

Fig 4. Same as Fig 3, except Day 2 outlook issued at 1230 pm Feb 4th. a) Categorical outlook,b) Probabilistic outlook

2

8/8/2019 sup-tues


of the elevated mixed layer (Carlson et al. 1983)and resultant cap would delay any surface-based,deep moist convection in association with thisfeature until later in the afternoon within the warmsector. This latter scenario would increase thelikelihood of a squall line being forced eastward bydeep convergence along the surface front duringthe evening and overnight. This would limitpotential to an isolated significant tornado threat,with a main threat of widespread damagingsurface winds. Model forecast soundings fromboth the NAM and Eta-KF control member fromthe SREF supported this scenario, as surfaceheating was forecast to be inadequate to break thecap until arrival of a strong surface cold front latein the day (Fig. 5).

In addition, some of the simulated radar reflectivity products available with the higher resolution WRF data sets (Weiss et al. 2006),tended to suggest rather weak and disorganized

convection over the lower Mississippi valleythrough the afternoon (Fig. 6). SPC forecastersare aware of the limitations with these explicit,convection-allowing models, particularly in lowinstability environments, due in part to the ~4 kmgrid length that is relatively coarse to resolveconvective-scale updrafts. However, operationaluse of high-resolution WRF output has provenuseful in similar situations by providing uniqueconvective details (e.g. mode and evolution).

While it was felt that the environment waspotentially supportive of a “High Risk”, thesecomplicating factors left enough uncertainty that a

Moderate Risk was maintained in the initial Day 1outlook issued at midnight CST 5 February.

These uncertainties persisted through theremainder of the night and into the next day. Theweak, southern stream short wave troughincreased deep moist convection in the form of elevated thunderstorms over north central Texas,which spread quickly into eastern Oklahoma andwestern Arkansas through the early morning hoursof the 5th. However, by daybreak it becameapparent the primary effects from this low latitudeshort wave trough would overspread the Ozarkregion, and likely not impact most of the warm

sector during the day. Modified forecastsoundings, using slightly warmer surfacetemperatures as predicted by local NWS forecastoffices, indicated capping would therefore likely bebroken across most of the warm sector by the midafternoon. In addition, a comparison of theobserved 12 UTC Little Rock, AR sounding themorning of the 5th with the 12 UTC 21 January1999 sounding (another cool season tornadooutbreak over the Mid South), indicated cap

Fig 5. Eta-kf 21 hr forecast valid at 00z Feb. 6 th

overlaid with observed sounding valid the sametime. Red- temperature and green- dew point for the forecast Eta-kf sounding. Obs. in purple.

Fig 6. 21 hour forecast of the 1 km agl simulated reflectivity from the 00z Feb 6 run of the WRF-NMM4.

Fig 7. Observed sounding valid at 12z for Feb.5 th, 2008 (red and green) and Jan. 21st , 1999(purple)

3

8/8/2019 sup-tues


erosion was farther advanced with weaker cappingin place on the morning of the 5th (Fig. 7). Thesefactors increased confidence that the severe threatwould evolve into a significant tornado outbreak,given the very favorable large scale environmentand the likelihood of discrete thunderstormdevelopment ahead of the main surface cold front.The subsequent Day 1 outlook issued at 7 amCST included a High Risk centered over the MidSouth region (Fig. 8).

Even after the introduction of the High Riskarea, uncertainty regarding the timing and locationof the greatest threat lingered well into the earlyafternoon. Some 12 UTC numerical modelguidance continued to suggest the development of widespread thunderstorms by late morning or early afternoon over the Arklatex region inassociation with the lead short wave trough liftingout of the lower Rio Grande Valley. However, asthe morning progressed, observational data trends

confirmed that the cap was holding across muchof the warm sector and significant, discretethunderstorm development would likely be delayeduntil peak heating that afternoon, but would stillprecede the main surface cold front. This allowedforecasters to discount the spurious numericalguidance and focus more intently on theobservations and short term model solutions thatstill remained plausible. As a result, the mid-morning convective outlook (issued at 1030 amCST) expanded the High Risk and moreaccurately captured the ensuing event. The firsttwo tornado watches for the outbreak were issued

shortly after 2 pm CST and 3 pm CST for the area,and both included the “Particularly DangerousSituation” wording that is reserved for significanttornado threats.

4. CONCLUSIONS

A commonly accepted scenario for high impactsevere weather events is that forecasters gainconfidence as model guidance and observed datagradually converge on a common solution thatincreases in accuracy as the event time nears.For example, it is assumed that the data will

gradually indicate that an increasing threat existsfor a high impact event, and forecasters begin toaddress the high end potential at an earlier point inthe forecast timeline. As additional data becomeavailable, specific details of the convectivescenario and better specification of the threat areaare provided in subsequent forecast issuances.

At other times, the data may begin to indicatethat a more limited threat exists than was earlier anticipated. In these instances, however, some

lower potential for a significant event typicallyremains, and the forecaster is faced with thedecision of how best to downplay the threatcontained in existing forecasts, while stillacknowledging that a lower level of threat mayexist. Typically, the use of probabilistic forecastproducts provide a vehicle to better reflectforecaster uncertainty (confidence), althoughinterpretation of the probabilistic information bymembers of the user community requiresadditional communication and education efforts.

In both of these examples, a consistent signalamongst the various observational and model datasets increases forecaster confidence in modifyingthe prevailing forecasts.

However, it is not uncommon for the variousdata sets (e.g., model guidance, satellite imagery,observed soundings, etc.) to contain mixed signalsregarding the likelihood of a high impact event,especially for smaller scale phenomena such as

Fig 8. Same as Fig 3, except Day 1 outlook issued at 1300z Feb 4th. a) Categorical outlook, b)Tornado probabilities

4

8/8/2019 sup-tues


severe thunderstorms and tornadoes wherepredictability on the storm scale is relatively low. Inthese instances, forecaster certainty may actuallydecrease with time as increasing volumes of newer and more detailed observational data andmodel guidance become available, often providinga variety of mesoscale and storm scale scenariosthat were not resolved by the larger scale data.Given the limitations in our ability to sample andresolve in sufficient detail the four-dimensionalstructure of the atmospheric in real time(especially the distribution of water vapor), anumber of the differing scenarios can all appear tobe plausible, and it can be very challenging todetermine which of the scenarios are most likely tooccur. Thus, even in situations leading up tosignificant severe weather events, lowpredictability on the smaller scales is inherentlypresent, and forecasters may become lessconfident in the impending threat as the event

nears. This is consistent with previous studies(e.g. Heideman et al. 1993) that showed that moredata by itself does not necessarily result inimproved forecasts, and more data may actuallycontribute to a reduction in forecast quality unlessbetter ways are found to incorporate the data intothe decision-making process.

We have used the Super Tuesday case toillustrate the latter forecast scenario. Despite thevery favorable large-scale environment, which wascorrectly predicted a week in advance, a highdegree of uncertainty persisted into the morning of 5 February concerning the evolution of the

mesoscale environment and subsequentconvective response. This uncertainty stemmedfrom conflicting short term model solutions,disparities between the model guidance andobservational data, and questions about how theenvironment would evolve in a very dynamicatmospheric pattern. But by “stepping back” andrefocusing on the overall synoptic scale pattern,and utilizing observational data to identifyimportant trends that provided insights into thetiming and location of when and where the capwas likely to break across the region, forecasterswere able to place a higher level of confidence in

predicting a significant tornado outbreak.

5. REFERENCES

Carlson, T.N., S.G. Benjamin, G.S. Forbes and Y.-F. Li, 1983: Elevated mixed layers in thesevere-storm environment – conceptual modeland case studies. Mon. Wea. Rev ., 111,1453-1473.

Doswell, C. A., III, S. J. Weiss, and R. H. Johns,1993: Tornado forecasting—A review. TheTornado: Its Structure, Dynamics, Predictionand Hazards, Geophys. Monogr., No. 79,Amer. Geophys. Union, 557–571.

Guyer, J.L., D.A. Imy, A. Kis, and K. Venable,2006: Cool Season Significant (F2-F5)Tornadoes in the Gulf Coast States. Preprints,23nd Conf. Severe Local Storms, St. Louis MO.

Heideman, K. F., T. R. Stewart, W. R. Moninger,and P. Reagan-Cirincione, 1993: The Weather Information and Skill Experiment (WISE): Theeffect of varying levels of information onforecast skill. Wea. Forecasting, 8, 25–36.

Weiss, S.J., D.R. Bright, J.S. Kain, J.J. Levit, M.E.Pyle, Z.I. Janjic, B.S. Ferrier, and J. Du, 2006:Complementary Use of Short-range Ensemble

and 4.5 KM WRF-NMM Model Guidance for Severe Weather Forecasting at the StormPrediction Center. Preprints, 23rd Conf. SevereLocal Storms, St. Louis MO.

5

sup-tues

Documents